CUP: JIT Comment UPdater

Directory Structure

baselines: the implementation of nnupdater and scripts for running baselines
models: neural models
scripts: the scripts for conducting experiments

Prepare Requirements

Java 8
Install python dependencies through conda
Install nlg-eval and setup

conda env create -f environment.yml

pip install git+https://github.com/Maluuba/nlg-eval.git@master
# set the data_path
nlg-eval --setup ${data_path}

Download Dataset

Our dataset, trained model and archived results can be downloaded from here
Another archive of this project can be found at https://tinyurl.com/jitcomment
By default, we store the dataset in ../dataset

Train

cd scripts
# 0 for GPU 0
./train_model.sh 0 CoAttnBPBAUpdater models.updater.CoAttnBPBAUpdater ../dataset

Infer and Evaluate

cd scripts
./infer_eval.sh 0 CoAttnBPBAUpdater models.updater.CoAttnBPBAUpdater ../dataset

Build Vocab Yourself

You can also build the vocabularies by yourself instead of using the one provided with our dataset.

# download fastText pre-trained model
cd ../dataset
wget https://dl.fbaipublicfiles.com/fasttext/vectors-crawl/cc.en.300.bin.gz
gunzip cc.en.300.bin.gz

cd scripts
./build_vocab.sh ../../dataset/cc.en.300.bin

Run Baselines

Clone FracoUpdater:

# clone FracoUpdater
git clone https://github.com/Tbabm/FracoUpdater

Install FracoUpdater's dependencies according to its README
Run

python -m baselines.run_baselines run_all_baselines

The results will be placed in the dataset directory, and can be evaluated using CUP/eval.py

Run CUP's Variants

cd scripts
# 0 for GPU 0
./run_variants.sh 0

Get Readable Result

python -m tools dump_all_readables

readable files can be found in results

megumigachi/CUP